git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From f83a275dbc5ca1721143698e844243fcadfabf6a Mon Sep 17 00:00:00 2001
   2 From: Mel Gorman <mel@csn.ul.ie>
   3 Date: Thu, 28 May 2009 14:34:40 -0700
   4 Subject: mm: account for MAP_SHARED mappings using VM_MAYSHARE and not VM_SHARED in hugetlbfs
   5
   6 From: Mel Gorman <mel@csn.ul.ie>
   7
   8 commit f83a275dbc5ca1721143698e844243fcadfabf6a upstream.
   9
  10 Addresses http://bugzilla.kernel.org/show_bug.cgi?id=13302
  11
  12 hugetlbfs reserves huge pages but does not fault them at mmap() time to
  13 ensure that future faults succeed.  The reservation behaviour differs
  14 depending on whether the mapping was mapped MAP_SHARED or MAP_PRIVATE.
  15 For MAP_SHARED mappings, hugepages are reserved when mmap() is first
  16 called and are tracked based on information associated with the inode.
  17 Other processes mapping MAP_SHARED use the same reservation.  MAP_PRIVATE
  18 track the reservations based on the VMA created as part of the mmap()
  19 operation.  Each process mapping MAP_PRIVATE must make its own
  20 reservation.
  21
  22 hugetlbfs currently checks if a VMA is MAP_SHARED with the VM_SHARED flag
  23 and not VM_MAYSHARE.  For file-backed mappings, such as hugetlbfs,
  24 VM_SHARED is set only if the mapping is MAP_SHARED and the file was opened
  25 read-write.  If a shared memory mapping was mapped shared-read-write for
  26 populating of data and mapped shared-read-only by other processes, then
  27 hugetlbfs would account for the mapping as if it was MAP_PRIVATE.  This
  28 causes processes to fail to map the file MAP_SHARED even though it should
  29 succeed as the reservation is there.
  30
  31 This patch alters mm/hugetlb.c and replaces VM_SHARED with VM_MAYSHARE
  32 when the intent of the code was to check whether the VMA was mapped
  33 MAP_SHARED or MAP_PRIVATE.
  34
  35 Signed-off-by: Mel Gorman <mel@csn.ul.ie>
  36 Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
  37 Cc: Ingo Molnar <mingo@elte.hu>
  38 Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
  39 Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
  40 Cc: <starlight@binnacle.cx>
  41 Cc: Eric B Munson <ebmunson@us.ibm.com>
  42 Cc: Adam Litke <agl@us.ibm.com>
  43 Cc: Andy Whitcroft <apw@canonical.com>
  44 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  45 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  46 Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
  47
  48 ---
  49  mm/hugetlb.c |   26 +++++++++++++-------------
  50  1 file changed, 13 insertions(+), 13 deletions(-)
  51
  52 --- a/mm/hugetlb.c
  53 +++ b/mm/hugetlb.c
  54 @@ -286,7 +286,7 @@ void resv_map_release(struct kref *ref)
  55  static struct resv_map *vma_resv_map(struct vm_area_struct *vma)
  56  {
  57         VM_BUG_ON(!is_vm_hugetlb_page(vma));
  58 -       if (!(vma->vm_flags & VM_SHARED))
  59 +       if (!(vma->vm_flags & VM_MAYSHARE))
  60                 return (struct resv_map *)(get_vma_private_data(vma) &
  61                                                         ~HPAGE_RESV_MASK);
  62         return 0;
  63 @@ -295,7 +295,7 @@ static struct resv_map *vma_resv_map(str
  64  static void set_vma_resv_map(struct vm_area_struct *vma, struct resv_map *map)
  65  {
  66         VM_BUG_ON(!is_vm_hugetlb_page(vma));
  67 -       VM_BUG_ON(vma->vm_flags & VM_SHARED);
  68 +       VM_BUG_ON(vma->vm_flags & VM_MAYSHARE);
  69
  70         set_vma_private_data(vma, (get_vma_private_data(vma) &
  71                                 HPAGE_RESV_MASK) | (unsigned long)map);
  72 @@ -304,7 +304,7 @@ static void set_vma_resv_map(struct vm_a
  73  static void set_vma_resv_flags(struct vm_area_struct *vma, unsigned long flags)
  74  {
  75         VM_BUG_ON(!is_vm_hugetlb_page(vma));
  76 -       VM_BUG_ON(vma->vm_flags & VM_SHARED);
  77 +       VM_BUG_ON(vma->vm_flags & VM_MAYSHARE);
  78
  79         set_vma_private_data(vma, get_vma_private_data(vma) | flags);
  80  }
  81 @@ -323,7 +323,7 @@ static void decrement_hugepage_resv_vma(
  82         if (vma->vm_flags & VM_NORESERVE)
  83                 return;
  84
  85 -       if (vma->vm_flags & VM_SHARED) {
  86 +       if (vma->vm_flags & VM_MAYSHARE) {
  87                 /* Shared mappings always use reserves */
  88                 h->resv_huge_pages--;
  89         } else if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
  90 @@ -339,14 +339,14 @@ static void decrement_hugepage_resv_vma(
  91  void reset_vma_resv_huge_pages(struct vm_area_struct *vma)
  92  {
  93         VM_BUG_ON(!is_vm_hugetlb_page(vma));
  94 -       if (!(vma->vm_flags & VM_SHARED))
  95 +       if (!(vma->vm_flags & VM_MAYSHARE))
  96                 vma->vm_private_data = (void *)0;
  97  }
  98
  99  /* Returns true if the VMA has associated reserve pages */
 100  static int vma_has_reserves(struct vm_area_struct *vma)
 101  {
 102 -       if (vma->vm_flags & VM_SHARED)
 103 +       if (vma->vm_flags & VM_MAYSHARE)
 104                 return 1;
 105         if (is_vma_resv_set(vma, HPAGE_RESV_OWNER))
 106                 return 1;
 107 @@ -890,7 +890,7 @@ static int vma_needs_reservation(struct
 108         struct address_space *mapping = vma->vm_file->f_mapping;
 109         struct inode *inode = mapping->host;
 110
 111 -       if (vma->vm_flags & VM_SHARED) {
 112 +       if (vma->vm_flags & VM_MAYSHARE) {
 113                 pgoff_t idx = vma_hugecache_offset(h, vma, addr);
 114                 return region_chg(&inode->i_mapping->private_list,
 115                                                         idx, idx + 1);
 116 @@ -915,7 +915,7 @@ static void vma_commit_reservation(struc
 117         struct address_space *mapping = vma->vm_file->f_mapping;
 118         struct inode *inode = mapping->host;
 119
 120 -       if (vma->vm_flags & VM_SHARED) {
 121 +       if (vma->vm_flags & VM_MAYSHARE) {
 122                 pgoff_t idx = vma_hugecache_offset(h, vma, addr);
 123                 region_add(&inode->i_mapping->private_list, idx, idx + 1);
 124
 125 @@ -1862,7 +1862,7 @@ retry_avoidcopy:
 126          * at the time of fork() could consume its reserves on COW instead
 127          * of the full address range.
 128          */
 129 -       if (!(vma->vm_flags & VM_SHARED) &&
 130 +       if (!(vma->vm_flags & VM_MAYSHARE) &&
 131                         is_vma_resv_set(vma, HPAGE_RESV_OWNER) &&
 132                         old_page != pagecache_page)
 133                 outside_reserve = 1;
 134 @@ -1969,7 +1969,7 @@ retry:
 135                 clear_huge_page(page, address, huge_page_size(h));
 136                 __SetPageUptodate(page);
 137
 138 -               if (vma->vm_flags & VM_SHARED) {
 139 +               if (vma->vm_flags & VM_MAYSHARE) {
 140                         int err;
 141                         struct inode *inode = mapping->host;
 142
 143 @@ -2073,7 +2073,7 @@ int hugetlb_fault(struct mm_struct *mm,
 144                         goto out_unlock;
 145                 }
 146
 147 -               if (!(vma->vm_flags & VM_SHARED))
 148 +               if (!(vma->vm_flags & VM_MAYSHARE))
 149                         pagecache_page = hugetlbfs_pagecache_page(h,
 150                                                                 vma, address);
 151         }
 152 @@ -2223,7 +2223,7 @@ int hugetlb_reserve_pages(struct inode *
 153          * to reserve the full area even if read-only as mprotect() may be
 154          * called to make the mapping read-write. Assume !vma is a shm mapping
 155          */
 156 -       if (!vma || vma->vm_flags & VM_SHARED)
 157 +       if (!vma || vma->vm_flags & VM_MAYSHARE)
 158                 chg = region_chg(&inode->i_mapping->private_list, from, to);
 159         else {
 160                 struct resv_map *resv_map = resv_map_alloc();
 161 @@ -2246,7 +2246,7 @@ int hugetlb_reserve_pages(struct inode *
 162                 hugetlb_put_quota(inode->i_mapping, chg);
 163                 return ret;
 164         }
 165 -       if (!vma || vma->vm_flags & VM_SHARED)
 166 +       if (!vma || vma->vm_flags & VM_MAYSHARE)
 167                 region_add(&inode->i_mapping->private_list, from, to);
 168         return 0;
 169  }